NovaData: Autonomous AI Data Scientist

🚀 Inspiration

Data science is often bogged down by the "80/20 rule"—where 80% of the time is spent on tedious data cleaning and only 20% on actual analysis. I was inspired to build NovaData to flip this script. I wanted to create an "Autonomous Scientist" that handles the grunt work, allowing users to move from a raw CSV to a predictive model in seconds.

🧠 How I Built It

NovaData is powered by Gemini 2.5 Flash. Unlike traditional chatbots, NovaData uses Native Code Execution to act as a reasoning agent.

The Brain: The LLM analyzes the data structure and writes real Python code.
The Sandbox: That code is executed in a secure environment to perform IQR-based outlier detection, data type casting, and statistical audits.
The Interface: Built with Streamlit, the dashboard provides a professional, dark-mode experience with real-time KPI metrics and natural language visualization.

🛠️ Challenges I Faced

The biggest technical hurdle was ensuring the agent could handle specific data formats and connection stability. I encountered a 400 MIME-type error during the file upload process, which I resolved by implementing a custom configuration for text/plain file handling. I also integrated high-timeout HTTP options to ensure large datasets (like the 500-entry socioeconomic data) could be processed without interruption.

📈 What I Learned

Building this project taught me the true power of Agentic Workflows. I learned that when an LLM is given tools—like a Python interpreter and Scikit-Learn—it transcends being a "text generator" and becomes a functional, high-precision tool for scientific discovery.

🔮 Future Scope

I plan to expand NovaData to support multi-file joins and automated hyperparameter tuning for more complex ML models like Random Forests and XGBoost.